Dynamic turbocompression, a century-old technology, has seen remarkable advancements, leveraging well-understood design fundamentals and physical phenomena to achieve impressive performance. However, this does not imply that all design rules are established or the technology is stagnant. Analogous to the enduring evolution of internal combustion engines (ICE), despite the emergence of new technologies like electric drivetrains and hydrogen fuel cells, ICEs have witnessed major advancements in power and efficiency. Similarly, centrifugal compressors continue to evolve, benefiting from advancements in peripheral technologies such as computer-aided engineering (CAE), multifidelity simulation, additive manufacturing, digital controls and electric drives, ensuring their relevance and ongoing development.
The modern compressor design process begins with a comprehensive analysis of the system, defining boundary specifications at nominal conditions and assessing the compressor’s full operating range. Since compressors handle fluids, creating an engineering model of the thermofluid system is crucial for determining the operating envelope and flow path design requirements. CAE tools are now widely accessible, facilitating the democratization of toolsets and enabling increases in productivity and reductions in design time and costs.
During the conceptual phase of design, employing reduced-order modeling techniques is crucial to simplify the system from a complex 3D problem to a manageable 0D or 1D problem. This involves defining reasonable component performances for simulating the thermofluid system using a lumped parameter approach. In this context, 0D and 1D refer to different aspects of the model. In a lumped parameter model, each component (e.g., compressor, heat exchanger, etc.) is described either by performance parameters (0D) or physical attributes (1D). Performance parameters encompass efficiencies, pinch temperatures and pressure drops, while physical attributes include surface areas, lengths and thermal conductivities. 1D models rely on physical principles and empirical or semi-empirical correlations that allow determination of performance parameters for each component. Certain software packages even enable the designer to have mixed 0D and 1D fidelity components in a single lumped parameter model.
During the initial stages of compressor design (or any turbomachine), it is vital to develop 0D models for system concepts, which are then refined to 1D to improve fidelity, accuracy and boundary condition definition for centrifugal compressors. In addition to these advantages, a system model may also enable the following benefits through the entire life cycle:
- Provide users early product specifications.
- Help determine compressor architecture and automation of the system predictions with automated data transfer between the compressor model and the system model.
- Incorporate advanced 2D and 3D predictive models of subcomponents into the system modeling for high fidelity system studies, commissioning support, continuous performance monitoring during duty and other digital twin features.
- Incorporate artificial intelligence (AI) methods to generate preliminary machine design geometry and performance predictions directly at the system level.
Having performed proper system design and analysis, it is time to move on to designing the compressor itself. Here, several multidisciplinary activities occur and can be divided into swimlanes.
1. Architecture and layout design: Conceptualize drive, bearing, thermal management and aerodynamics, iterating between system and aerodynamic design.
2. Aerodynamic flow path design: Design fluid handling surfaces and blade shapes based on boundary conditions.
3. Drive system design/selection: Consider drive system interfaces with aerodynamic design for speed, torque and power limitations.
4. Mechanical design: Analyze flow path components for structure and modal frequencies; then design housing, assembly and mechanical auxiliaries.
5. Rotordynamics analysis and bearing design/selection: Ensure frequency separation margin between operating and resonance frequencies, iteratively coupled with bearing system design and interface with architecture/layout and aerodynamics.
Traditionally, architecture selection was accomplished using heuristic methods, and the design process advanced linearly to aerodynamics, mechanical design and rotordynamics and bearing design. Any issues discovered during this process, leading to non-feasibility, would necessitate extensive iterations back to nearly the starting point. Thanks to the prevalence of fast-solving analysis and automated design tools, these steps can be taken almost in parallel.
Now, one of the core disciplines of compressor design, the aerodynamics, is practiced. Here, modern design methodology is to employ generative design. It lets designers provide boundary conditions to computational algorithms, while the algorithms use those conditions in conjunction with a design space boundary to output design geometry variants. The process is enabled by the calculation methods developed throughout a century of centrifugal compressor knowledge combined with modern order-reduction techniques and computational automation capabilities.
These advances allow creating hundreds, if not thousands, of design candidates in the time it takes to make a cup of coffee. Each design is quickly evaluated using a reduced-order analysis technique for basic performance predictions. By using this technique, the designer can promptly perform various trade-off analyses on the generative design dataset(s) and interface with the drive system design engineers, rotordynamics and other disciplines to eliminate candidates or entire parts of the design space that are incompatible with the other disciplines.
To use a concrete example, suppose a centrifugal compressor is developed to be driven by an electric motor and variable frequency drive (VFD). The rotational speed is open to selection, depending on the capabilities of different drive systems selections. The generative design technique enables the designer to identify viable geometries and operating speeds and down-select not only individual candidates, but entire regions of a design space as suitable or unsuitable, while evaluating multiple additional performance metrics such as efficiency and operating range.
Generative design enables quick optimization of the architecture and basic geometry, but it is only the beginning of aerodynamic design. The optimum design candidate should now be evaluated using reduced-order aerodynamic modeling, which may involve kinematic analysis, meanline solver coupled with loss models and 2D throughflow solvers in order to assess flow path performance under various operating conditions. It is prudent to use such tools to develop preliminary performance curves.
With modern algorithms, such models can also be coupled to the system analysis code and develop accurate system-level predictions of operation and performance. Furthermore, if the generation of performance prediction curves is sufficiently fast, they will offer valuable feedback for tuning the blade profile and meridional curve shapes when optimizing the aerodynamic flow path.
At this point in the aerodynamic design, the design is mature enough that other swimlane disciplines should begin. Among them are the mechanical design and critically important rotordynamics and bearing design. When the mass and material properties of the rotating flow path components are well defined, and once the drive system is at least conceptually understood, an initial rotordynamics analysis should be performed to understand potential problems with resonance and critical vibrating modes.
Advanced 3D flow analysis, as is done in computational fluid dynamics (CFD), is the standard these days. At a minimum, a CFD analysis should be performed to validate early predictions performed during the design process. Where optimum performance is critical, CFD should be used in conjunction with advanced optimization algorithms that may include gradient-based methods, genetic or evolutionary algorithms and possibly even machine learning (ML) methods to incorporate surrogate predictive capabilities, as was demonstrated by Goldenberg, et al.1
Modern integrated design environments utilize turbomachinery geometry methods. They can either integrate 3D finite element analysis (FEA) and CFD methods directly into the development environment or have developed, integrated and automated geometry and data transfer methods to various commercial analysis packages. Having designed the aerodynamic flow path, data about geometry and boundary conditions needs to be transferred to an FEA solver to perform static structural and modal/harmonic analyses on a solid body model. Additionally, CFD validation is done to improve accuracy of earlier performance predictions and carry out further optimizations.
References
Goldenberg, V., Gorman, J.M., Simon, T. and Sparrow, E.M., 2020, September. A Numerical Approach to Centrifugal Compressor Stage Flow Path Design Synthesis and Optimization. In Turbo Expo: Power for Land, Sea, and Air (Vol. 84089, p. V02CT35A034). American Society of Mechanical Engineers.
